Adding helpers to parse Bigtable create cluster operation.

[dhermes]

A CreateCluster request response doesn't actually indicate
success or failure. Rather it returns a Cluster message with
the validated request parts inside and a current_operation
attached.

We implement _process_operation so that we can determine the
ID of that long-running operation (so it can be checked for
completion / success, if desired by the user). In addition
we seek to notify the user when the request began.

From the service definition we know that the current_operation
is a long-running operation and that:

The embedded operation's "metadata" field type is CreateClusterMetadata,
The embedded operation's "response" field type is Cluster, if successful.

The Operation metadata is of type Any (which uses a type_url
and raw bytes to provide any protobuf message type in a single
field, but still allow it to be parsed into it's true type after
the fact). So we expect CreateCluster responses to have long-running
operations with a type URL matching CreateClusterMetadata.

As a result, we introduce a utility (_parse_pb_any_to_native) for
parsing an Any field into the native protobuf type specified by the
type URL. Since we know we need to handle CreateClusterMetadata values,
we add a default mapping (_TYPE_URL_MAP) from the corresponding type url
for that message type to the native Python type.

The CreateClusterMetadata type has request_time and
finish_time fields of type Timestamp so we also
add the _pb_timestamp_to_datetime helper for converting protobuf
messages into native Python datetime.datetime objects.

[dhermes]

@tseaver Sorry this is kind of a doozy. I wrote a long description to try to give you some context and quick windows into what the actual data returned from the API is.

---

This PR is doing something very simple: parse the information from the CreateCluster response. But it takes a bit of machinery for doing this.

[tseaver]

LGTM